AOMedia AV1 Codec
ratectrl.h
1 /*
2  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #ifndef AOM_AV1_ENCODER_RATECTRL_H_
13 #define AOM_AV1_ENCODER_RATECTRL_H_
14 
15 #include "aom/aom_codec.h"
16 #include "aom/aom_integer.h"
17 
18 #include "aom_ports/mem.h"
19 
20 #include "av1/common/av1_common_int.h"
21 #include "av1/common/blockd.h"
22 
23 #ifdef __cplusplus
24 extern "C" {
25 #endif
26 
29 // Bits Per MB at different Q (Multiplied by 512)
30 #define BPER_MB_NORMBITS 9
31 
32 // Use this macro to turn on/off use of alt-refs in one-pass mode.
33 #define USE_ALTREF_FOR_ONE_PASS 1
34 
35 // Threshold used to define if a KF group is static (e.g. a slide show).
36 // Essentially, this means that no frame in the group has more than 1% of MBs
37 // that are not marked as coded with 0,0 motion in the first pass.
38 #define STATIC_KF_GROUP_THRESH 99
39 #define STATIC_KF_GROUP_FLOAT_THRESH 0.99
40 
41 // The maximum duration of a GF group that is static (e.g. a slide show).
42 #define MAX_STATIC_GF_GROUP_LENGTH 250
43 
44 #define MIN_GF_INTERVAL 4
45 #define MAX_GF_INTERVAL 32
46 #define FIXED_GF_INTERVAL 16
47 #define MAX_GF_LENGTH_LAP 16
48 
49 #define FIXED_GF_INTERVAL_RT 80
50 #define MAX_GF_INTERVAL_RT 160
51 
52 #define MAX_NUM_GF_INTERVALS 15
53 
54 #define MAX_ARF_LAYERS 6
55 // #define STRICT_RC
56 
57 #define DEFAULT_KF_BOOST_RT 2300
58 #define DEFAULT_GF_BOOST_RT 2000
59 
60 // A passive rate control strategy for screen content type in real-time mode.
61 // When it is turned on, the compression performance is improved by
62 // 7.8% (overall_psnr), 5.0% (VMAF) on average. Some clips see gains
63 // over 20% on metric.
64 // The downside is that it does not guarantee frame size.
65 // Since RT mode has a tight restriction on buffer overflow control, we
66 // turn it off by default.
67 #define RT_PASSIVE_STRATEGY 0
68 #define MAX_Q_HISTORY 1000
69 
70 typedef struct {
71  int resize_width;
72  int resize_height;
73  uint8_t superres_denom;
74 } size_params_type;
75 
76 enum {
77  INTER_NORMAL,
78  GF_ARF_LOW,
79  GF_ARF_STD,
80  KF_STD,
81  RATE_FACTOR_LEVELS
82 } UENUM1BYTE(RATE_FACTOR_LEVEL);
83 
84 enum {
85  KF_UPDATE,
86  LF_UPDATE,
87  GF_UPDATE,
88  ARF_UPDATE,
89  OVERLAY_UPDATE,
90  INTNL_OVERLAY_UPDATE, // Internal Overlay Frame
91  INTNL_ARF_UPDATE, // Internal Altref Frame
92  FRAME_UPDATE_TYPES
93 } UENUM1BYTE(FRAME_UPDATE_TYPE);
94 
95 enum {
96  REFBUF_RESET, // Clear reference frame buffer
97  REFBUF_UPDATE, // Refresh reference frame buffer
98  REFBUF_STATES
99 } UENUM1BYTE(REFBUF_STATE);
100 
101 typedef enum {
102  NO_RESIZE = 0,
103  DOWN_THREEFOUR = 1, // From orig to 3/4.
104  DOWN_ONEHALF = 2, // From orig or 3/4 to 1/2.
105  UP_THREEFOUR = -1, // From 1/2 to 3/4.
106  UP_ORIG = -2, // From 1/2 or 3/4 to orig.
107 } RESIZE_ACTION;
108 
109 typedef enum { ORIG = 0, THREE_QUARTER = 1, ONE_HALF = 2 } RESIZE_STATE;
110 
111 #define MAX_FIRSTPASS_ANALYSIS_FRAMES 150
112 typedef enum region_types {
113  STABLE_REGION = 0,
114  HIGH_VAR_REGION = 1,
115  SCENECUT_REGION = 2,
116  BLENDING_REGION = 3,
117 } REGION_TYPES;
118 
119 typedef struct regions {
120  int start;
121  int last;
122  double avg_noise_var;
123  double avg_cor_coeff;
124  double avg_sr_fr_ratio;
125  double avg_intra_err;
126  double avg_coded_err;
127  REGION_TYPES type;
128 } REGIONS;
129 
134 typedef struct {
135  // Rate targetting variables
136 
145  int this_frame_target; // Actual frame target after rc adjustment.
146 
151 
156 
161 
166 
171 
176 
178  int min_gf_interval;
179  int max_gf_interval;
180  int static_scene_max_gf_interval;
187  int frames_since_key;
188  int frames_to_fwd_kf;
189  int is_src_frame_alt_ref;
190  int sframe_due;
191 
192  int high_source_sad;
193  uint64_t avg_source_sad;
194  uint64_t prev_avg_source_sad;
195  uint64_t frame_source_sad;
196 
197  int avg_frame_bandwidth; // Average frame size target for clip
198  int min_frame_bandwidth; // Minimum allocation used for any frame
199  int max_frame_bandwidth; // Maximum burst rate allowed for a frame.
200  int prev_avg_frame_bandwidth;
201 
202  int ni_av_qi;
203  int ni_tot_qi;
204 
205  int decimation_factor;
206  int decimation_count;
207 
217 
220  // rate control history for last frame(1) and the frame before(2).
221  // -1: overshoot
222  // 1: undershoot
223  // 0: not initialized.
224  int rc_1_frame;
225  int rc_2_frame;
226  int q_1_frame;
227  int q_2_frame;
228 
234 
236  // Track amount of low motion in scene
237  int avg_frame_low_motion;
238  int cnt_zeromv;
239 
240  // signals if number of blocks with motion is high
241  int percent_blocks_with_motion;
242 
243  // Maximum value of source sad across all blocks of frame.
244  uint64_t max_block_source_sad;
245 
246  // For dynamic resize, 1 pass cbr.
247  RESIZE_STATE resize_state;
248  int resize_avg_qp;
249  int resize_buffer_underflow;
250  int resize_count;
251 
252  // Flag to disable content related qp adjustment.
253  int rtc_external_ratectrl;
254 
255  // Stores fast_extra_bits of the current frame.
256  int frame_level_fast_extra_bits;
257 
258  double frame_level_rate_correction_factors[RATE_FACTOR_LEVELS];
259 
260  int frame_num_last_gf_refresh;
261 
262  int prev_coded_width;
263  int prev_coded_height;
265 } RATE_CONTROL;
266 
270 typedef struct {
271  // Sub-gop level Rate targetting variables
272 
276  int64_t gf_group_bits;
277 
281  int kf_boost;
282 
287 
291  int gf_intervals[MAX_NUM_GF_INTERVALS];
292 
297 
299  int num_regions;
300 
301  REGIONS regions[MAX_FIRSTPASS_ANALYSIS_FRAMES];
302  int regions_offset; // offset of regions from the last keyframe
303  int frames_till_regions_update;
304 
305  int baseline_gf_interval;
306 
307  int constrained_gf_group;
308 
309  int this_key_frame_forced;
310 
311  int next_key_frame_forced;
318 
323 
328 
332  int arf_q;
333 
335  float_t arf_boost_factor;
336 
337  int base_layer_qp;
338 
339  // Total number of stats used only for kf_boost calculation.
340  int num_stats_used_for_kf_boost;
341 
342  // Total number of stats used only for gfu_boost calculation.
343  int num_stats_used_for_gfu_boost;
344 
345  // Total number of stats required by gfu_boost calculation.
346  int num_stats_required_for_gfu_boost;
347 
348  int enable_scenecut_detection;
349 
350  int use_arf_in_this_kf_group;
351 
352  int ni_frames;
353 
354  double tot_q;
361 
365  int avg_frame_qindex[FRAME_TYPES];
366 
367 #if CONFIG_FPMT_TEST
372  int temp_active_best_quality[MAX_ARF_LAYERS + 1];
373 
378  int temp_last_boosted_qindex;
379 
384  double temp_avg_q;
385 
390  int temp_last_q[FRAME_TYPES];
391 
396  int temp_projected_frame_size;
397 
402  int64_t temp_total_actual_bits;
403 
408  int64_t temp_buffer_level;
409 
414  int64_t temp_vbr_bits_off_target;
415 
420  int64_t temp_vbr_bits_off_target_fast;
421 
426  double temp_rate_correction_factors[RATE_FACTOR_LEVELS];
427 
432  int temp_rate_error_estimate;
433 
438  int temp_rolling_arf_group_target_bits;
439 
444  int temp_rolling_arf_group_actual_bits;
445 
450  int64_t temp_bits_left;
451 
456  int temp_extend_minq;
457 
462  int temp_extend_maxq;
463 
468  int temp_extend_minq_fast;
469 #endif
473  int active_best_quality[MAX_ARF_LAYERS + 1];
474 
479 
483  double avg_q;
484 
488  int last_q[FRAME_TYPES];
489 
494  double rate_correction_factors[RATE_FACTOR_LEVELS];
495 
500 
505 
509  int64_t buffer_level;
510 
515 
520 
525 
531 
536 
542 
547  int q_history[MAX_Q_HISTORY];
549 
552 struct AV1_COMP;
553 struct AV1EncoderConfig;
554 struct GF_GROUP;
555 
556 void av1_primary_rc_init(const struct AV1EncoderConfig *oxcf,
557  PRIMARY_RATE_CONTROL *p_rc);
558 
559 void av1_rc_init(const struct AV1EncoderConfig *oxcf, RATE_CONTROL *rc);
560 
561 int av1_estimate_bits_at_q(const struct AV1_COMP *cpi, int q,
562  double correction_factor);
563 
564 double av1_convert_qindex_to_q(int qindex, aom_bit_depth_t bit_depth);
565 
566 void av1_rc_init_minq_luts(void);
567 
568 int av1_rc_get_default_min_gf_interval(int width, int height, double framerate);
569 // Note av1_rc_get_default_max_gf_interval() requires the min_gf_interval to
570 // be passed in to ensure that the max_gf_interval returned is at least as bis
571 // as that.
572 int av1_rc_get_default_max_gf_interval(double framerate, int min_gf_interval);
573 
574 // Generally at the high level, the following flow is expected
575 // to be enforced for rate control:
576 // First call per frame, one of:
577 // av1_get_one_pass_rt_params()
578 // av1_get_second_pass_params()
579 // depending on the usage to set the rate control encode parameters desired.
580 //
581 // Then, call encode_frame_to_data_rate() to perform the
582 // actual encode. This function will in turn call encode_frame()
583 // one or more times, followed by:
584 // av1_rc_postencode_update_drop_frame()
585 //
586 // The majority of rate control parameters are only expected
587 // to be set in the av1_get_..._params() functions and
588 // updated during the av1_rc_postencode_update...() functions.
589 // The only exceptions are av1_rc_drop_frame() and
590 // av1_rc_update_rate_correction_factors() functions.
591 
592 // Functions to set parameters for encoding before the actual
593 // encode_frame_to_data_rate() function.
594 struct EncodeFrameInput;
595 
596 // Post encode update of the rate control parameters based
597 // on bytes used
598 void av1_rc_postencode_update(struct AV1_COMP *cpi, uint64_t bytes_used);
599 // Post encode update of the rate control parameters for dropped frames
600 void av1_rc_postencode_update_drop_frame(struct AV1_COMP *cpi);
601 
617  int is_encode_stage, int width,
618  int height);
621 // Decide if we should drop this frame: For 1-pass CBR.
622 // Changes only the decimation count in the rate control structure
623 int av1_rc_drop_frame(struct AV1_COMP *cpi);
624 
625 // Computes frame size bounds.
626 void av1_rc_compute_frame_size_bounds(const struct AV1_COMP *cpi,
627  int this_frame_target,
628  int *frame_under_shoot_limit,
629  int *frame_over_shoot_limit);
630 
645 int av1_rc_pick_q_and_bounds(struct AV1_COMP *cpi, int width, int height,
646  int gf_index, int *bottom_index, int *top_index);
647 
660 int av1_rc_regulate_q(const struct AV1_COMP *cpi, int target_bits_per_frame,
661  int active_best_quality, int active_worst_quality,
662  int width, int height);
663 
665 // Gets the appropriate bpmb ennumerator based on the frame and content type
666 int av1_get_bpmb_enumerator(FRAME_TYPE frame_type,
667  const int is_screen_content_type);
668 
669 // Estimates bits per mb for a given qindex and correction factor.
670 int av1_rc_bits_per_mb(const struct AV1_COMP *cpi, FRAME_TYPE frame_type,
671  int qindex, double correction_factor,
672  int accurate_estimate);
673 
674 // Clamping utilities for bitrate targets for iframes and pframes.
675 int av1_rc_clamp_iframe_target_size(const struct AV1_COMP *const cpi,
676  int64_t target);
677 int av1_rc_clamp_pframe_target_size(const struct AV1_COMP *const cpi,
678  int target, uint8_t frame_update_type);
679 
680 // Find q_index corresponding to desired_q, within [best_qindex, worst_qindex].
681 // To be precise, 'q_index' is the smallest integer, for which the corresponding
682 // q >= desired_q.
683 // If no such q index is found, returns 'worst_qindex'.
684 int av1_find_qindex(double desired_q, aom_bit_depth_t bit_depth,
685  int best_qindex, int worst_qindex);
686 
687 // Computes a q delta (in "q index" terms) to get from a starting q value
688 // to a target q value
689 int av1_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget,
690  aom_bit_depth_t bit_depth);
691 
692 // Computes a q delta (in "q index" terms) to get from a starting q value
693 // to a value that should equate to the given rate ratio.
694 int av1_compute_qdelta_by_rate(const struct AV1_COMP *cpi,
695  FRAME_TYPE frame_type, int qindex,
696  double rate_target_ratio);
697 
698 int av1_frame_type_qdelta(const struct AV1_COMP *cpi, int q);
699 
700 void av1_rc_update_framerate(struct AV1_COMP *cpi, int width, int height);
701 
702 void av1_rc_set_gf_interval_range(const struct AV1_COMP *const cpi,
703  RATE_CONTROL *const rc);
704 
705 void av1_set_target_rate(struct AV1_COMP *cpi, int width, int height);
706 
707 int av1_resize_one_pass_cbr(struct AV1_COMP *cpi);
708 
709 void av1_rc_set_frame_target(struct AV1_COMP *cpi, int target, int width,
710  int height);
711 
712 void av1_adjust_gf_refresh_qp_one_pass_rt(struct AV1_COMP *cpi);
713 
715  int gf_update);
716 
730  const struct AV1_COMP *const cpi, FRAME_UPDATE_TYPE frame_update_type);
731 
743 
756  const struct AV1_COMP *cpi, FRAME_UPDATE_TYPE frame_update_type);
757 
769 
789  FRAME_TYPE *const frame_type,
790  const struct EncodeFrameInput *frame_input,
791  unsigned int frame_flags);
792 
807 int av1_encodedframe_overshoot_cbr(struct AV1_COMP *cpi, int *q);
808 
820 int av1_q_mode_get_q_index(int base_q_index, int gf_update_type,
821  int gf_pyramid_level, int arf_q);
822 
832 int av1_get_arf_q_index(int base_q_index, int gfu_boost, int bit_depth,
833  double arf_boost_factor);
834 
835 #if !CONFIG_REALTIME_ONLY
836 struct TplDepFrame;
844 int av1_get_arf_q_index_q_mode(struct AV1_COMP *cpi,
845  struct TplDepFrame *tpl_frame);
846 #endif
847 #ifdef __cplusplus
848 } // extern "C"
849 #endif
850 
851 #endif // AOM_AV1_ENCODER_RATECTRL_H_
Describes the codec algorithm interface to applications.
enum aom_bit_depth aom_bit_depth_t
Bit depth for codecThis enumeration determines the bit depth of the codec.
int av1_calc_pframe_target_size_one_pass_vbr(const struct AV1_COMP *const cpi, FRAME_UPDATE_TYPE frame_update_type)
Calculates how many bits to use for a P frame in one pass vbr.
void av1_rc_update_rate_correction_factors(struct AV1_COMP *cpi, int is_encode_stage, int width, int height)
Updates the rate correction factor linking Q to output bits.
Definition: ratectrl.c:689
int av1_encodedframe_overshoot_cbr(struct AV1_COMP *cpi, int *q)
Increase q on expected encoder overshoot, for CBR mode.
Definition: ratectrl.c:3339
void av1_get_one_pass_rt_params(struct AV1_COMP *cpi, FRAME_TYPE *const frame_type, const struct EncodeFrameInput *frame_input, unsigned int frame_flags)
Setup the rate control parameters for 1 pass real-time mode.
int av1_calc_pframe_target_size_one_pass_cbr(const struct AV1_COMP *cpi, FRAME_UPDATE_TYPE frame_update_type)
Calculates how many bits to use for a P frame in one pass cbr.
int av1_rc_pick_q_and_bounds(struct AV1_COMP *cpi, int width, int height, int gf_index, int *bottom_index, int *top_index)
Picks q and q bounds given the rate control parameters in cpi->rc.
Definition: ratectrl.c:2054
int av1_calc_iframe_target_size_one_pass_cbr(const struct AV1_COMP *cpi)
Calculates how many bits to use for an i frame in one pass cbr.
int av1_calc_iframe_target_size_one_pass_vbr(const struct AV1_COMP *const cpi)
Calculates how many bits to use for an i frame in one pass vbr.
int av1_rc_regulate_q(const struct AV1_COMP *cpi, int target_bits_per_frame, int active_best_quality, int active_worst_quality, int width, int height)
Estimates q to achieve a target bits per frame.
void av1_set_rtc_reference_structure_one_layer(AV1_COMP *cpi, int gf_update)
Setup the reference prediction structure for 1 pass real-time.
Definition: ratectrl.c:2750
Main encoder configuration data structure.
Definition: encoder.h:915
Top level encoder structure.
Definition: encoder.h:2759
Input frames and last input frame.
Definition: encoder.h:3470
Data related to the current GF/ARF group and the individual frames within the group.
Definition: firstpass.h:344
Primary Rate Control parameters and status.
Definition: ratectrl.h:270
int64_t bits_off_target
Definition: ratectrl.h:530
int rate_error_estimate
Definition: ratectrl.h:514
double avg_q
Definition: ratectrl.h:483
int64_t maximum_buffer_size
Definition: ratectrl.h:327
int kf_boost
Definition: ratectrl.h:281
int64_t starting_buffer_level
Definition: ratectrl.h:317
int64_t total_actual_bits
Definition: ratectrl.h:499
int64_t vbr_bits_off_target_fast
Definition: ratectrl.h:524
int rolling_target_bits
Definition: ratectrl.h:535
int64_t buffer_level
Definition: ratectrl.h:509
int gfu_boost
Definition: ratectrl.h:286
int64_t optimal_buffer_level
Definition: ratectrl.h:322
int arf_q
Definition: ratectrl.h:332
int rolling_actual_bits
Definition: ratectrl.h:541
int last_boosted_qindex
Definition: ratectrl.h:478
int cur_gf_index
Definition: ratectrl.h:296
int64_t total_target_bits
Definition: ratectrl.h:504
int last_kf_qindex
Definition: ratectrl.h:360
int64_t gf_group_bits
Definition: ratectrl.h:276
int64_t vbr_bits_off_target
Definition: ratectrl.h:519
Rate Control parameters and status.
Definition: ratectrl.h:134
int best_quality
Definition: ratectrl.h:216
int intervals_till_gf_calculate_due
Definition: ratectrl.h:175
int this_frame_target
Definition: ratectrl.h:145
int frames_to_key
Definition: ratectrl.h:185
int base_frame_target
Definition: ratectrl.h:141
int projected_frame_size
Definition: ratectrl.h:150
int worst_quality
Definition: ratectrl.h:212
int sb64_target_rate
Definition: ratectrl.h:160
int coefficient_size
Definition: ratectrl.h:155
int active_worst_quality
Definition: ratectrl.h:233
int frames_since_golden
Definition: ratectrl.h:165
int frames_till_gf_update_due
Definition: ratectrl.h:170